Identifying Humans by Their Walk and Generating New Motions Using Hidden Markov Models
|
|
- Coleen Little
- 5 years ago
- Views:
Transcription
1 Identifying Humans by Their Walk and Generating New Motions Using Hidden Markov Models CPSC 532A Topics in AI: Graphical Models and CPSC 526 Computer Animation December 15, 2004 Andrew Adam Saleema Amershi 1. INTRODUCTION The problems we consider in this work are that of recognizing a human figure from 3D motion capture data of a person walking, and also of generating new movement data useful for movies, simulations or games. What is it that could make a human distinguishable from another by their walk? The answer is gait. Previous work in this area has shown that if all points on the body are considered it is possible to differentiate between individuals. We attempt to recognize people by modeling each individual s gait using a Hidden Markov Model (HMM). The HMM is a good choice for modeling a walk cycle because it can model sequential processes. The data that we consider comes from the Carnergie Mellon University (CMU) motion capture database, but in theory any data that drives or represents a human figure could be used. We build each HMM by converting motion capture data from.amc format to Matlab matrices (Converter), using feature vectors from that data to produce cluster sets (Clusterer), initializing the model with appropriate values based on the cluster set (Initializer) and then improving the model parameters to best represent the original data (Trainer). Once we have an HMM for each person, we can make use of the model in two applications: identifying individuals from new unseen data (Identifier) or from training data so as to test the model s accuracy (Tester), and generating new pieces of motion (Generator). The rest of this paper is organized as follows: Section 2 discusses our work in the context of previous work on motion recognition and generation. In Section 3 we introduce and describe our approach to recognition and generation. Section 4 gives an overview of our results (further results can be examined by running our matlab code and viewing our previously generated files). In Section 5 we discuss the limitations of our system and describe avenues for future extension. 2. BACKGROUND and RELATED WORK Distinguishing people by the way they walk is part of a greater problem of recognizing people using different types of biometrics. The background for our idea comes from Kale et al. [7] where silhouettes of a human figure are extracted from video images and used to build an HMM and then to recognize the person. This system would be useful where fingerprinting or eye scanning is unsuitable, and where there is surveillance cameras or with night vision goggles. Unfortunately its accuracy depends on the style of clothing the person wears. Instead of using a silhouette we will use 3D motion capture data but apply HMM s in a similar way. HMMs appear in many previous works, Rabiners paper [15] describing their use in speech recognition being a classic example. This pape r discusses what an HMM is and why they are useful and goes on to state that such models are rich in mathematical structure
2 and hence can form the theoretical basis for use in a wide range of applications. [12] also tries to extend the HMM framework for the application of recognizing speech. Alongside this, [3] is a presentation on dimensionality reduction and feature selection and how they work, and also the application to human tracking. There are many other works that attempt recognition from 2D or 3D data. [18] uses video data and HMMs to try and recognize which sign language symbol is being used at a particular moment in time. They manage to recognize 99% of the symbols when colored gloves are worn and 92% without gloves, from a 40 word lexicon. [9] attempts to add HMMs when recognizing activities that have a predefined context and inherent semantics, such as in the card game Blackjack. They extract behavior and strategies in real-time. [21] attempts to decode different types of gestures from video, using a technique developed by the same authors in [22] called a Parametric HMM (PHMM), that is used to model parameterized gestures (gestures that show a meaningful variation). An example of this would be a pointing gesture where the important parameter is direction, and therefore would be parameterized by the Cartesian coordinates that indicate direction. The PHMM recognizes the gesture and also estimates the quantifying parameters using the Expectation-Maximization algorithm (EM). [16] also attempts classification from video sequences, and uses mixed discrete/continuous states to couple perception with classification. A spline contour is used to track the contour of the person, along with mixed state condensation filtering. Another work uses a switching linear dynamic system (SLDS) to analyze and track a human figure [13] cast in the framework of Dynamic Bayesian Networks (DBN s) [11]. [2] uses HMM s on video and the EM algorithm for human gait recognition and [14] is a more narrow recognition problem which attempts to recognize types of movements such as tennis stroke, again using an HMM. [20] is a Masters thesis by Donald O. Tanguay Jr. at MIT, which recognizes different types of motion gestures in video. [23] presents a 3D recognition scheme and efficiently computes the 3D appearance using a region-based coarse stereo matching algorithm. An unsupervised learning scheme is carried out to capture the cluster structure of these feature volumes, then the image sequence of a gesture is converted to a sequence of symbols that initialize the cluster identities of each image pair. Two schemes are used (forward HMM s and Neural Networks) to model the dynamics of the gestures, and they achieve a recognition accuracy of 96%. [5] creates what the authors call a Continuous Human Movement Recognition framework, which uses multiple HMM s to infer the movement skill from an alphabet of dynemes (units of full body skills.) Also used is a clone-body-model which is dynamically sized and texture mapped for more robust tracking of humans on video of both edges and texture regions. There are also many examples of work that generate new movement data from existing data. [1] is an example of stylistic motion synthesis that uses learning across a varied set of choreographed dancing data to create new dancing motions. They achieve this using a Stylistic HMM (SHMM), which can also be driven by video, scripts or even by noise to generate new choreography and synthesize virtual motion capture in many different styles. [4] classifies human movement compactly by using primitives, by filtering, segmenting and applying the Principle Component Algorithm (PCA) but in this case uses it for generating movement in robotics. [6] synthesizes a walking human motion that follows a sample trajectory and also generates a synthetic partner for a dancer who is acquired through motion capture. [19] synthesizes novel motion sequences from a database of motion capture examples, providing the flexibility of keyframe animation with the realism of motion capture data. Finally, [17] constructs new parametric models out of captured motion capture data, using a multi-step approximation for the interpolation function and motion data compressed by PCA, for use in real-time applications. They can vary the model by age, height, weight and gender. 3. SYSTEM OVERVIEW We aim to model an individual s walk using a Hidden Markov Model (HMM) with discrete hidden states and continuous observations from the states. An HMM can model a walk cycle because it can model sequential stochastic processes, or states, where the
3 probability of a state depends on previous states. For a walk cycle we use a left-right HMM, where a state can only be reached by itself or by a single other state. The states are hidden and one can only observe a sequence of observations generated from a sequence of states. In our case we observe skeletal joint configurations of a person, and assume they are generated from the states by a Gaussian probability density function. That is we assume that at each consecutive time step a new skeletal pose is generated from some state. Figure 1 shows an example cyclic left-right HMM. q 0 x 0 q 1 x 1 Figure 1. Cyclic Left-Right Hidden Markov Model In Figure 1, the q i nodes represent hid den states and the shaded x i nodes the observations from these states. An HMM is parameterized by: transition probabilities a ij, 1 i,j N, which are the probabilities of transitioning from state q i to state q j in successive time slices, initial probabilities π i, 1 i N, which are the probabilities of a sequence starting in any state q i, and observation probabilities b i (x), 1 i N, which are q 2 x 2 q 3 x 3 the probabilities of observing x given being in state q i. These parameters can be summarized by λ = (p, A, B) where p is a vector of probabilities and A is a matrix of probabilities. For our system we will be modeling continuous Gaussian observations and so we can replace B with a Gaussian probability density function N(x; m, S ) where m and S are the mean and covariance of the distribution. Our goal is to automatically learn the parameters λ = (p, A, N) from motion capture data of walking sequences. Each person s walk will be modeled by one identically structured HMM. The unique parameters learned will be used to recognize people from new observation sequences. By performing random walks on the HMM we will also be able to produce new walking motions for each person. An overview of our system is shown in Figure 2. In the rest of this section we will explain our approach by describing each of the modules in Figure 2 in further detail. 3.1 Identification Converting For our system we use Acclaim formatted motion capture data from the CMU motion capture data base. The Acclaim format consists of i) a.asf file specifying skeleton information, and ii) a.amc file containing motion information for a sequence of frames. Each frame specifies the position and orientation of the root of the skeleton (6 dimensions) and joint angles for each joint in a skeleton (56 dimensions). For our system, we exclude skeletal information and ignore the root position and orientation, and focus recognition based on how a person is moving using only the joint angle information in the amc files. All of the individuals we use have structurally similar skeletons. Figure 2. System Overview
4 We extract a 56 dimensional feature vector for every frame in a sequence and then preprocess the data by centering the feature vectors about the mean pose and normalizing Clustering After extracting the necessary features, we identify clusters of data which will correspond to the states in our HMM. Each feature vector can be considered as a point in a 56 dimensional joint space. And a walk can be considered as cycles through similar poses. These similar poses (or points near each other in the joint space) can then be grouped into clusters, so as to partition the joint space. To find these data clusters we employ the K-means clustering algorithm. The K-means algorithm identifies k vectors, each corresponding to the mean of a cluster in the joint space. Given a number, k, of clusters to find, the K-means algorithm iteratively identifies cluster means, mi for i = 1,..,k, and assigns data points, x n, to clusters. We initialize the algorithm by randomly selecting feature vectors from our data to be the cluster means. For each feature vector, the squared distance between the feature and each cluster mean ( x n - m i 2 ) is then computed, and the feature vector is assigned to the cluster which minimizes the distance to its mean. After each feature vector in the data is assigned to a cluster, the algorithm recomputes the cluster means: m i = 1/N i Σ x n. The algorithm then reassigns the data to the new clusters (represented by the new means) and so on for a maximum of 100 iterations. For each HMM we identify four clusters using K- means. We found that this is an adequate and natural representation for a walk cycle in our system. Figure 3 shows an example of the joint configurations corresponding to the four cluster means found for a person by our system Intializing We can now initialize the parameters, λ = (p, A, N), of our HMM. Recall that we want to model the joint configurations of a walking sequence as continuous observations from discrete states. We model these continuous observations using: b i (x) = N(x; m i, s i ), 1 i N where b i (x) models the probability density function of observation x generated from state i as a Gaussian. The four partitions (N = 4) of the feature vectors found by K-means are used to initialize the Gaussian parameters m and s (the mean vector and covariance matrix respectively). The identified cluster means correspond directly to the mean vectors m i. The form of s i is spherical where the covariance is set to be the squared distance to the next closest mean vector. We evenly distribute probability to the priors p of the HMM. This means that we assume it is equally likely Right Foot Up Right Foot Forward Left Foot Up Left Foot Forward Figure 3. Four Cluster Means Found Using K-Means
5 to begin in any state of our HMM. EM (see Section 3.1.4) will adjust this prior to reflect the actual initial state. The transition matrix A specifying the probability of transitioning from any state to any other state, is initialized based on frequency counts from the feature vector data. Each feature vector corresponds to a single frame in a sequence. From this and the partition information found in the last section, we can count how many times we transition from one state to another in consecutive frames. That is the element a ij (1 i,j N) in A estimates the probability p(q t+1 j q t i ) of transitioning from state q i to state q j in consecutive frames by counting the number of times in the sequence that we transition from state q i to q j, divided by the total number of times we are in state q i Training To improve our initial model parameters λ = (p, A, N), we use the expectation-maximization (EM) algorithm. Our goal is to maximize the complete log likelihood p(x, q λ) that a sequence X of data vectors x was generated by our HMM by adjusting the HMM parameters λ. However, we do not have complete data because the state variables q i are hidden. So we must maximize (the M step) the expected complete log likelihood argmax λ Σ q p(q X, λ)log p(x, q λ) with respect to the parameters λ, where p(q X, λ) is estimated in the E step. That is, in the E step we calculate the sufficient statistics for the HMM parameters λ using the forward-backwards algorithm, and then in the M step we compute the maximum likelihood (ML) estimates of the parameters. These ML parameters are the improved HMM parameters. We can then repe at the E and M steps using the new parameters to improve our model Testing/Identifying We can create and train an HMM for several different people using the approach described in Sections for each individual. Once we have a set of HMMs, recognition of a person from a new sequence of observations x amounts to calculating the log likelihood of x given each of the models ( log p(x λ ) for each HMM). The model that gives the highest probability is the model that most likely generated the new sequence x. To calculate p(x λ) we can marginalize out the hidden states q from the joint distribution p( x, q λ ). That is, the likelihood is the sum of the likelihoods of x for every possible sequence of states q in the model log p(x λ ) = Σ q log p( x, q λ). Using the forwards step of the forwards backwards algorithm, we can efficiently compute this probability. In the forwards algorithm we can accumulate or store probabilities of sequences up to any time step t. Then we can update the probability for the next time step t+1 by multiplying the previously stored probabilities with the next transition and observation probabilities. The probabilities we store are α t (i) = p (x 1, x 2, x t, q t = q i λ) which are the probabilities of observing the sequence x up to time step t and ending up in state q i given the model. At each successive time step we need only to update α by α t+1 (j) = [Σ i α t (i) a ij ] b j (x t+1) where we sum the first equation for α t (i) over all possible end states q i multiplied by the transition probabilities of going from that end state to the current state q j. We then only need to multiply by the probability of observing the current observation x from state q j. This will in effect compute the likelihood of the data for an HMM. We compute the log likelihood of the data for each HMM in our set and then identify a person by choosing the HMM that produced the highest value. We use this same procedure for testing each HMM against the training data (the data that was used to create and train the HMM) and for recognizing new data. 3.2 Motion Generation For motion generation, a trained HMM for an individual is taken and the means of the clustered states are retrieved. Each HMM was built with four states which resulted in two mean vectors with outstretched legs and two with (roughly) straight legs, which is how we expected it (see Figure 3). To create a new walk we worked out the number of frames to generate by using the values in the transition
6 matrix A. The diagonal probability values in the transition matrix are large (~0.97), which correspond to staying in the current state for a certain length of time. Then in each row, there is one small probability (~0.03), the index of which corresponds to which unique state follows the current state. The two remaining values are zero to signify that the other states are impossible as followers (as necessary for the left-right HMM). By dividing one by the small value we get a figure (~35) which tells us the number of frames there should be between transitions. Also the order of the states is deduced from the indices of the small values: to to to to 2 State Order = We average the number of frames between transitions to work out approximately how many frames there should be between each state. This number of frames is created by linearly interpolating between each mean state vector. The difference between the values of each state is worked out and divided by the number of transitions to create an alpha value which can be added to each frame to create a new corresponding position. Also we have to reinsert position data that we removed for the identification process. We choose a start position and add the number of frames in a single walk cycle to that position in the test sequence to see how far the character has traveled, and we use this to interpolate between the start and end of our new sequence points. 4. RESULTS We ran our system using four different test subjects, person 7, person 35, person 39 and person 44 from the CMU database as they had lots of walking sequences to work with. Each HMM was created and trained using three walking sequenc es from the subject. We tested each HMM against these training sequences and also against four unseen sequences per subject. We tested two different types of covariance for the clusters, spherical and PPCA, and set the number of clusters to four. The results for each of the different types of covariance were different. Spherical achieved 6 out of 12 correct when tested against the training data. PPCA got three wrong therefore 9 out of 12. Spherical got correct classification of 10 out of 16 of the unseen files, and PPCA got 13 out of 16. So, while it seems PPCA is better, further tests may reveal that each one is better with a certain style of walk. In terms of generation we took the HMM for each subject and then ran our generator on them, producing one walk cycle for each of the four subjects DISCUSSION AND FUTURE WORK In our system we used a 56 dimensional feature vector for our HMM observations. This is a large amount of data which may contain much redundant information. To minimize redundancy and compress the data we could have used Principal Component Analysis (PCA) to extract the significant components of the data. PCA is a common technique used for compression that computes the eigenvectors of the high dimensional space. By projecting our high dimensional data onto a lower dimensional space spanned by some number of eigenvectors, n, corresponding to the n largest eigenvalues found, we could have reduced this dimensionality and still retain the significant information needed for recognition. For generation we would need to invert this reduction to project back to our high dimensional space and retrieve the new motion data. However, the high dimensionality features used in our system did not substantially reduce performance and so we left this out. Some joint angles contain more significant information about a walking motion. For example, the hip rotation may contain more variation then the ankle rotations when a person walks. In the future we could add weighting information to the feature vectors to give more importance to these more significant joints. 1 To view our results, see our code and generated motions at
7 We used a standard left-right HMM for our system in which the state durations are modeled by a probability p i (d) = (a ii ) d-1 (1- a ii ), where a ii is the self transition probability for state q i, and d is the number of consecutive self transitions. A variable duration HMM in which we explicitly specify the state durations would more naturally represent a walking motion because this would allow us to explicitly state the duration that an HMM must be in a certain state before transitioning to the next. That is, we do not want to jump from one state to the next without generating an appropriate number of observations from one state. In the future we could implement this type of HMM and compare performance to our current system. The generated output that we produced could be improved in many ways. The generated motion is not ideal because we get a lot of foot sliding and floating figures. We could have interpolated the position data in a better fashio n to keep the new walk along the floor. We also could have eliminated foot-sliding by using inverse kinematics and constraints to force the characters feet to be planted when in contact with the ground so as to have the character move in a more realistic manner such as in [8]. With more time we would make these additions to our system. In theory our system could also be used for recognizing different types of motion rather than recognizing different people from the same type of motion. However this was not tested, but could easily be added as an extension in the future. We also could extend our system to be used in conjunction with vision systems that can track human figures in video sequences. For example [10], takes several 2D views of the human body, locates joint positions, and then uses them to estimate the body configuration and pose in 3D space. This appears to be an effective way of pulling 3D data out of 2D video so perhaps it, or similar systems, could be used as input to our system in the future. 6. REFERENCES [1] Brand, M. and Hertzmann, A. Style Machines. SIGGRAPH 2000, ACM Press (2000), [2] Bregler, C. Learning and Recognizing Human Dynamics in Video Sequences. Proc. IEEE Conf. Comp. Vision and Pattern Recognition 1997, 7 Pages. [3] Cohen, I. et al. Feature Selection Using Principle Feature Analysis. Slide Presentation. [4] Fod, A., Mataric, M.J. and Jenkins O.C. Automated Derivation of Primitives for Movement Classification. Autonomous Robots 12, 1 (2002), [5] Green, R. and Guan, L. Quantifying and Recognizing Human Movement Patterns from Monocular Video Images - Part I: A New Framework for Modeling Human Motion. IEEE Trans. on Circuits and Systems for Video Technology 14, 2 (2004), [6] Hsu, E., Gentry, S. and Popovic, J. Example- Based Control of Human Motion. SIGGRAPH/Eurographics Syp. on Computer Animation 2004, ACM Press (2004), [7] Kale, A., Cuntoor, N. and Kruger, V. Gait-based Recognition of Humans Using Continuous HMMs. IEEE Int. Conf. on Automatic Face and Gesture Recognition 2002, IEEE Computer Society (2002), 336. [8] Kovar, L., Gleicher, M. and Schreiner J. Footskate Cleanup for Motion Capture Editing. SIGGRAPH 2003, ACM Press (2002), [9] Moore, D. and Essa, I. Recognizing Multitasked Activities using Stochastic Context-Free Grammar. SIGART 2002, AAAI (2002), [10] Mori, G. and Malik, J. Estimating Human Body Configurations using Shape Context Matching. ECCV 2002, [11] Murphy, K. Dynamic Bayesian Networks [12] Ostendorf, M., Digalakis, V. and Kimbal O. From HMMs to Segment Models: A Unified View of Stochastic Modelling for Speech Recognition. IEEE Trans. Speech Audio Process 4, 5 (1996), [13] Pavlovic, V. and Rehg, J. Impact of Dynamic Model Learning on Classification of Human Motion. CVPR 2000, Pages [14] Petkovic, M., Jonker, W. and Zivokic, Z. Recognizing Strokes in Tennis Videos Using Hidden Markov Models. Proc. of Int.
8 Conference on Visualization Imaging, and Image Processing [15] Rabiner, I. A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition. Proc. of IEEE 77, 2 (1989), [16] Rittscher, J., Blake, A., and Roberts, S. J. Towards the automatic analysis of complex human body motions. IVC(20) 12, (2002) [17] Soo Lim, I. and Thalmann, O. D11: Generative Models from Capture Motion. [18] Starner, T. and Pentland, A. Real-Time American Sign Language Recognition from Video Using Hidden Markov Models. SCV95, 1995 [19] Tanco, I. and Hilton, A. Realistic Synthesis of Novel Human Movements from a Database of Motion Capture Examples, Proc. of the IEEE Workshop on Human Motion HUMO [20] Tanguay D. Jr. Hidden Markov Models for Gesture Recognition, Masters Thesis at Massachusetts Institute of Technology, Cambridge, June 1993 [21] Wilson A. and Bobick, A. Nonlinear PHMMs for the Interpretation of Parameterized Gesture. Proc of the IEEE Conference on Computer Vision and Pattern Recognition 1998, Page 879. [22] Wilson, A. and Bobick, A. Parametric Hidden Markov Models for Gesture Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence 21, 9 (1999). [23] Ye, G., Corso, J. and Hager, G. Gesture Recognition Using 3D Appearance and Motion Features, 2004.
Epitomic Analysis of Human Motion
Epitomic Analysis of Human Motion Wooyoung Kim James M. Rehg Department of Computer Science Georgia Institute of Technology Atlanta, GA 30332 {wooyoung, rehg}@cc.gatech.edu Abstract Epitomic analysis is
More informationGenerating Different Realistic Humanoid Motion
Generating Different Realistic Humanoid Motion Zhenbo Li,2,3, Yu Deng,2,3, and Hua Li,2,3 Key Lab. of Computer System and Architecture, Institute of Computing Technology, Chinese Academy of Sciences, Beijing
More informationMotion Interpretation and Synthesis by ICA
Motion Interpretation and Synthesis by ICA Renqiang Min Department of Computer Science, University of Toronto, 1 King s College Road, Toronto, ON M5S3G4, Canada Abstract. It is known that high-dimensional
More informationHuman Motion Synthesis by Motion Manifold Learning and Motion Primitive Segmentation
Human Motion Synthesis by Motion Manifold Learning and Motion Primitive Segmentation Chan-Su Lee and Ahmed Elgammal Rutgers University, Piscataway, NJ, USA {chansu, elgammal}@cs.rutgers.edu Abstract. We
More informationUnsupervised Learning
Unsupervised Learning Learning without Class Labels (or correct outputs) Density Estimation Learn P(X) given training data for X Clustering Partition data into clusters Dimensionality Reduction Discover
More informationTerm Project Final Report for CPSC526 Statistical Models of Poses Using Inverse Kinematics
Term Project Final Report for CPSC526 Statistical Models of Poses Using Inverse Kinematics Department of Computer Science The University of British Columbia duanx@cs.ubc.ca, lili1987@cs.ubc.ca Abstract
More informationExpanding gait identification methods from straight to curved trajectories
Expanding gait identification methods from straight to curved trajectories Yumi Iwashita, Ryo Kurazume Kyushu University 744 Motooka Nishi-ku Fukuoka, Japan yumi@ieee.org Abstract Conventional methods
More information22 October, 2012 MVA ENS Cachan. Lecture 5: Introduction to generative models Iasonas Kokkinos
Machine Learning for Computer Vision 1 22 October, 2012 MVA ENS Cachan Lecture 5: Introduction to generative models Iasonas Kokkinos Iasonas.kokkinos@ecp.fr Center for Visual Computing Ecole Centrale Paris
More informationLearnt Inverse Kinematics for Animation Synthesis
VVG (5) (Editors) Inverse Kinematics for Animation Synthesis Anonymous Abstract Existing work on animation synthesis can be roughly split into two approaches, those that combine segments of motion capture
More informationRealistic Synthesis of Novel Human Movements from a Database of Motion Capture Examples
Realistic Synthesis of Novel Human Movements from a Database of Motion Capture Examples Appeared in Proceedings of the IEEE Workshop on Human Motion HUMO 2000 Luis Molina Tanco Adrian Hilton Centre for
More informationClustering K-means. Machine Learning CSEP546 Carlos Guestrin University of Washington February 18, Carlos Guestrin
Clustering K-means Machine Learning CSEP546 Carlos Guestrin University of Washington February 18, 2014 Carlos Guestrin 2005-2014 1 Clustering images Set of Images [Goldberger et al.] Carlos Guestrin 2005-2014
More informationHuman Upper Body Pose Estimation in Static Images
1. Research Team Human Upper Body Pose Estimation in Static Images Project Leader: Graduate Students: Prof. Isaac Cohen, Computer Science Mun Wai Lee 2. Statement of Project Goals This goal of this project
More informationImage Coding with Active Appearance Models
Image Coding with Active Appearance Models Simon Baker, Iain Matthews, and Jeff Schneider CMU-RI-TR-03-13 The Robotics Institute Carnegie Mellon University Abstract Image coding is the task of representing
More informationAnalyzing and Segmenting Finger Gestures in Meaningful Phases
2014 11th International Conference on Computer Graphics, Imaging and Visualization Analyzing and Segmenting Finger Gestures in Meaningful Phases Christos Mousas Paul Newbury Dept. of Informatics University
More informationLast week. Multi-Frame Structure from Motion: Multi-View Stereo. Unknown camera viewpoints
Last week Multi-Frame Structure from Motion: Multi-View Stereo Unknown camera viewpoints Last week PCA Today Recognition Today Recognition Recognition problems What is it? Object detection Who is it? Recognizing
More informationNOVEL HYBRID GENETIC ALGORITHM WITH HMM BASED IRIS RECOGNITION
NOVEL HYBRID GENETIC ALGORITHM WITH HMM BASED IRIS RECOGNITION * Prof. Dr. Ban Ahmed Mitras ** Ammar Saad Abdul-Jabbar * Dept. of Operation Research & Intelligent Techniques ** Dept. of Mathematics. College
More informationHAND-GESTURE BASED FILM RESTORATION
HAND-GESTURE BASED FILM RESTORATION Attila Licsár University of Veszprém, Department of Image Processing and Neurocomputing,H-8200 Veszprém, Egyetem u. 0, Hungary Email: licsara@freemail.hu Tamás Szirányi
More informationMotion Synthesis and Editing. Yisheng Chen
Motion Synthesis and Editing Yisheng Chen Overview Data driven motion synthesis automatically generate motion from a motion capture database, offline or interactive User inputs Large, high-dimensional
More informationFrom Gaze to Focus of Attention
From Gaze to Focus of Attention Rainer Stiefelhagen, Michael Finke, Jie Yang, Alex Waibel stiefel@ira.uka.de, finkem@cs.cmu.edu, yang+@cs.cmu.edu, ahw@cs.cmu.edu Interactive Systems Laboratories University
More informationA Dynamic Human Model using Hybrid 2D-3D Representations in Hierarchical PCA Space
A Dynamic Human Model using Hybrid 2D-3D Representations in Hierarchical PCA Space Eng-Jon Ong and Shaogang Gong Department of Computer Science, Queen Mary and Westfield College, London E1 4NS, UK fongej
More informationLearning Switching Linear Models of Human Motion
To appear in Neural Information Processing Systems 13, MIT Press, Cambridge, MA Learning Switching Linear Models of Human Motion Vladimir Pavlović and James M. Rehg Compaq - Cambridge Research Lab Cambridge,
More informationAutomatic Gait Recognition. - Karthik Sridharan
Automatic Gait Recognition - Karthik Sridharan Gait as a Biometric Gait A person s manner of walking Webster Definition It is a non-contact, unobtrusive, perceivable at a distance and hard to disguise
More informationDynamic Time Warping for Binocular Hand Tracking and Reconstruction
Dynamic Time Warping for Binocular Hand Tracking and Reconstruction Javier Romero, Danica Kragic Ville Kyrki Antonis Argyros CAS-CVAP-CSC Dept. of Information Technology Institute of Computer Science KTH,
More informationCS325 Artificial Intelligence Ch. 20 Unsupervised Machine Learning
CS325 Artificial Intelligence Cengiz Spring 2013 Unsupervised Learning Missing teacher No labels, y Just input data, x What can you learn with it? Unsupervised Learning Missing teacher No labels, y Just
More informationHuman Action Recognition Using Independent Component Analysis
Human Action Recognition Using Independent Component Analysis Masaki Yamazaki, Yen-Wei Chen and Gang Xu Department of Media echnology Ritsumeikan University 1-1-1 Nojihigashi, Kusatsu, Shiga, 525-8577,
More informationClustering: Classic Methods and Modern Views
Clustering: Classic Methods and Modern Views Marina Meilă University of Washington mmp@stat.washington.edu June 22, 2015 Lorentz Center Workshop on Clusters, Games and Axioms Outline Paradigms for clustering
More informationGraph-based High Level Motion Segmentation using Normalized Cuts
Graph-based High Level Motion Segmentation using Normalized Cuts Sungju Yun, Anjin Park and Keechul Jung Abstract Motion capture devices have been utilized in producing several contents, such as movies
More informationCombined Shape Analysis of Human Poses and Motion Units for Action Segmentation and Recognition
Combined Shape Analysis of Human Poses and Motion Units for Action Segmentation and Recognition Maxime Devanne 1,2, Hazem Wannous 1, Stefano Berretti 2, Pietro Pala 2, Mohamed Daoudi 1, and Alberto Del
More informationSwitching Hypothesized Measurements: A Dynamic Model with Applications to Occlusion Adaptive Joint Tracking
Switching Hypothesized Measurements: A Dynamic Model with Applications to Occlusion Adaptive Joint Tracking Yang Wang Tele Tan Institute for Infocomm Research, Singapore {ywang, telctan}@i2r.a-star.edu.sg
More informationFace Recognition Using Vector Quantization Histogram and Support Vector Machine Classifier Rong-sheng LI, Fei-fei LEE *, Yan YAN and Qiu CHEN
2016 International Conference on Artificial Intelligence: Techniques and Applications (AITA 2016) ISBN: 978-1-60595-389-2 Face Recognition Using Vector Quantization Histogram and Support Vector Machine
More informationLinear Discriminant Analysis in Ottoman Alphabet Character Recognition
Linear Discriminant Analysis in Ottoman Alphabet Character Recognition ZEYNEB KURT, H. IREM TURKMEN, M. ELIF KARSLIGIL Department of Computer Engineering, Yildiz Technical University, 34349 Besiktas /
More informationApplications Video Surveillance (On-line or off-line)
Face Face Recognition: Dimensionality Reduction Biometrics CSE 190-a Lecture 12 CSE190a Fall 06 CSE190a Fall 06 Face Recognition Face is the most common biometric used by humans Applications range from
More informationTime series, HMMs, Kalman Filters
Classic HMM tutorial see class website: *L. R. Rabiner, "A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition," Proc. of the IEEE, Vol.77, No.2, pp.257--286, 1989. Time series,
More information10-701/15-781, Fall 2006, Final
-7/-78, Fall 6, Final Dec, :pm-8:pm There are 9 questions in this exam ( pages including this cover sheet). If you need more room to work out your answer to a question, use the back of the page and clearly
More information9.913 Pattern Recognition for Vision. Class I - Overview. Instructors: B. Heisele, Y. Ivanov, T. Poggio
9.913 Class I - Overview Instructors: B. Heisele, Y. Ivanov, T. Poggio TOC Administrivia Problems of Computer Vision and Pattern Recognition Overview of classes Quick review of Matlab Administrivia Instructors:
More informationAn Introduction to Pattern Recognition
An Introduction to Pattern Recognition Speaker : Wei lun Chao Advisor : Prof. Jian-jiun Ding DISP Lab Graduate Institute of Communication Engineering 1 Abstract Not a new research field Wide range included
More informationNote Set 4: Finite Mixture Models and the EM Algorithm
Note Set 4: Finite Mixture Models and the EM Algorithm Padhraic Smyth, Department of Computer Science University of California, Irvine Finite Mixture Models A finite mixture model with K components, for
More informationRecognition: Face Recognition. Linda Shapiro EE/CSE 576
Recognition: Face Recognition Linda Shapiro EE/CSE 576 1 Face recognition: once you ve detected and cropped a face, try to recognize it Detection Recognition Sally 2 Face recognition: overview Typical
More informationGraph-Based Action Models for Human Motion Classification
Graph-Based Action Models for Human Motion Classification Felix Endres Jürgen Hess Wolfram Burgard University of Freiburg, Dept. of Computer Science, Freiburg, Germany Abstract Recognizing human actions
More informationRandom projection for non-gaussian mixture models
Random projection for non-gaussian mixture models Győző Gidófalvi Department of Computer Science and Engineering University of California, San Diego La Jolla, CA 92037 gyozo@cs.ucsd.edu Abstract Recently,
More informationThiruvarangan Ramaraj CS525 Graphics & Scientific Visualization Spring 2007, Presentation I, February 28 th 2007, 14:10 15:00. Topic (Research Paper):
Thiruvarangan Ramaraj CS525 Graphics & Scientific Visualization Spring 2007, Presentation I, February 28 th 2007, 14:10 15:00 Topic (Research Paper): Jinxian Chai and Jessica K. Hodgins, Performance Animation
More informationClustering & Dimensionality Reduction. 273A Intro Machine Learning
Clustering & Dimensionality Reduction 273A Intro Machine Learning What is Unsupervised Learning? In supervised learning we were given attributes & targets (e.g. class labels). In unsupervised learning
More informationIndividual Recognition Using Gait Energy Image
Individual Recognition Using Gait Energy Image Ju Han and Bir Bhanu Center for Research in Intelligent Systems University of California, Riverside, California 92521, USA jhan,bhanu @cris.ucr.edu Abstract
More informationModelStructureSelection&TrainingAlgorithmsfor an HMMGesture Recognition System
ModelStructureSelection&TrainingAlgorithmsfor an HMMGesture Recognition System Nianjun Liu, Brian C. Lovell, Peter J. Kootsookos, and Richard I.A. Davis Intelligent Real-Time Imaging and Sensing (IRIS)
More informationLearning Deformations of Human Arm Movement to Adapt to Environmental Constraints
Learning Deformations of Human Arm Movement to Adapt to Environmental Constraints Stephan Al-Zubi and Gerald Sommer Cognitive Systems, Christian Albrechts University, Kiel, Germany Abstract. We propose
More informationMonocular Human Motion Capture with a Mixture of Regressors. Ankur Agarwal and Bill Triggs GRAVIR-INRIA-CNRS, Grenoble, France
Monocular Human Motion Capture with a Mixture of Regressors Ankur Agarwal and Bill Triggs GRAVIR-INRIA-CNRS, Grenoble, France IEEE Workshop on Vision for Human-Computer Interaction, 21 June 2005 Visual
More informationA Performance Evaluation of HMM and DTW for Gesture Recognition
A Performance Evaluation of HMM and DTW for Gesture Recognition Josep Maria Carmona and Joan Climent Barcelona Tech (UPC), Spain Abstract. It is unclear whether Hidden Markov Models (HMMs) or Dynamic Time
More information3D Human Motion Analysis and Manifolds
D E P A R T M E N T O F C O M P U T E R S C I E N C E U N I V E R S I T Y O F C O P E N H A G E N 3D Human Motion Analysis and Manifolds Kim Steenstrup Pedersen DIKU Image group and E-Science center Motivation
More informationShort Survey on Static Hand Gesture Recognition
Short Survey on Static Hand Gesture Recognition Huu-Hung Huynh University of Science and Technology The University of Danang, Vietnam Duc-Hoang Vo University of Science and Technology The University of
More informationMotion Texture. Harriet Pashley Advisor: Yanxi Liu Ph.D. Student: James Hays. 1. Introduction
Motion Texture Harriet Pashley Advisor: Yanxi Liu Ph.D. Student: James Hays 1. Introduction Motion capture data is often used in movies and video games because it is able to realistically depict human
More informationHidden Markov Model for Sequential Data
Hidden Markov Model for Sequential Data Dr.-Ing. Michelle Karg mekarg@uwaterloo.ca Electrical and Computer Engineering Cheriton School of Computer Science Sequential Data Measurement of time series: Example:
More informationTai Chi Motion Recognition Using Wearable Sensors and Hidden Markov Model Method
Tai Chi Motion Recognition Using Wearable Sensors and Hidden Markov Model Method Dennis Majoe 1, Lars Widmer 1, Philip Tschiemer 1 and Jürg Gutknecht 1, 1 Computer Systems Institute, ETH Zurich, Switzerland
More informationAnimation Lecture 10 Slide Fall 2003
Animation Lecture 10 Slide 1 6.837 Fall 2003 Conventional Animation Draw each frame of the animation great control tedious Reduce burden with cel animation layer keyframe inbetween cel panoramas (Disney
More informationECE521: Week 11, Lecture March 2017: HMM learning/inference. With thanks to Russ Salakhutdinov
ECE521: Week 11, Lecture 20 27 March 2017: HMM learning/inference With thanks to Russ Salakhutdinov Examples of other perspectives Murphy 17.4 End of Russell & Norvig 15.2 (Artificial Intelligence: A Modern
More informationA Generative Model for Motion Synthesis and Blending Using Probability Density Estimation
A Generative Model for Motion Synthesis and Blending Using Probability Density Estimation Dumebi Okwechime and Richard Bowden University of Surrey, Guildford, Surrey, GU2 7XH, UK {d.okwechime,r.bowden}@surrey.ac.uk
More informationMachine Learning. B. Unsupervised Learning B.1 Cluster Analysis. Lars Schmidt-Thieme, Nicolas Schilling
Machine Learning B. Unsupervised Learning B.1 Cluster Analysis Lars Schmidt-Thieme, Nicolas Schilling Information Systems and Machine Learning Lab (ISMLL) Institute for Computer Science University of Hildesheim,
More informationA Unified Spatio-Temporal Articulated Model for Tracking
A Unified Spatio-Temporal Articulated Model for Tracking Xiangyang Lan Daniel P. Huttenlocher {xylan,dph}@cs.cornell.edu Cornell University Ithaca NY 14853 Abstract Tracking articulated objects in image
More informationDynamic Human Shape Description and Characterization
Dynamic Human Shape Description and Characterization Z. Cheng*, S. Mosher, Jeanne Smith H. Cheng, and K. Robinette Infoscitex Corporation, Dayton, Ohio, USA 711 th Human Performance Wing, Air Force Research
More informationNonlinear Image Interpolation using Manifold Learning
Nonlinear Image Interpolation using Manifold Learning Christoph Bregler Computer Science Division University of California Berkeley, CA 94720 bregler@cs.berkeley.edu Stephen M. Omohundro'" Int. Computer
More informationMotion Control with Strokes
Motion Control with Strokes Masaki Oshita Kyushu Institute of Technology oshita@ces.kyutech.ac.jp Figure 1: Examples of stroke-based motion control. Input strokes (above) and generated motions (below).
More informationHuman Detection and Tracking for Video Surveillance: A Cognitive Science Approach
Human Detection and Tracking for Video Surveillance: A Cognitive Science Approach Vandit Gajjar gajjar.vandit.381@ldce.ac.in Ayesha Gurnani gurnani.ayesha.52@ldce.ac.in Yash Khandhediya khandhediya.yash.364@ldce.ac.in
More informationRecognition Rate. 90 S 90 W 90 R Segment Length T
Human Action Recognition By Sequence of Movelet Codewords Xiaolin Feng y Pietro Perona yz y California Institute of Technology, 36-93, Pasadena, CA 925, USA z Universit a dipadova, Italy fxlfeng,peronag@vision.caltech.edu
More informationCS 534: Computer Vision Segmentation and Perceptual Grouping
CS 534: Computer Vision Segmentation and Perceptual Grouping Ahmed Elgammal Dept of Computer Science CS 534 Segmentation - 1 Outlines Mid-level vision What is segmentation Perceptual Grouping Segmentation
More informationTopics in AI (CPSC 532L): Multimodal Learning with Vision, Language and Sound. Lecture 12: Deep Reinforcement Learning
Topics in AI (CPSC 532L): Multimodal Learning with Vision, Language and Sound Lecture 12: Deep Reinforcement Learning Types of Learning Supervised training Learning from the teacher Training data includes
More informationHidden Markov Models. Slides adapted from Joyce Ho, David Sontag, Geoffrey Hinton, Eric Xing, and Nicholas Ruozzi
Hidden Markov Models Slides adapted from Joyce Ho, David Sontag, Geoffrey Hinton, Eric Xing, and Nicholas Ruozzi Sequential Data Time-series: Stock market, weather, speech, video Ordered: Text, genes Sequential
More informationGRAPH-BASED APPROACH FOR MOTION CAPTURE DATA REPRESENTATION AND ANALYSIS. Jiun-Yu Kao, Antonio Ortega, Shrikanth S. Narayanan
GRAPH-BASED APPROACH FOR MOTION CAPTURE DATA REPRESENTATION AND ANALYSIS Jiun-Yu Kao, Antonio Ortega, Shrikanth S. Narayanan University of Southern California Department of Electrical Engineering ABSTRACT
More informationDimension Reduction CS534
Dimension Reduction CS534 Why dimension reduction? High dimensionality large number of features E.g., documents represented by thousands of words, millions of bigrams Images represented by thousands of
More informationM I RA Lab. Speech Animation. Where do we stand today? Speech Animation : Hierarchy. What are the technologies?
MIRALab Where Research means Creativity Where do we stand today? M I RA Lab Nadia Magnenat-Thalmann MIRALab, University of Geneva thalmann@miralab.unige.ch Video Input (face) Audio Input (speech) FAP Extraction
More informationThesis Proposal : Switching Linear Dynamic Systems with Higher-order Temporal Structure. Sang Min Oh
Thesis Proposal : Switching Linear Dynamic Systems with Higher-order Temporal Structure Sang Min Oh sangmin@cc.gatech.edu 28th May 2008 Contents 1 Introduction 1 1.1 Automated Temporal Sequence Analysis.................................
More informationIntroduction to Pattern Recognition Part II. Selim Aksoy Bilkent University Department of Computer Engineering
Introduction to Pattern Recognition Part II Selim Aksoy Bilkent University Department of Computer Engineering saksoy@cs.bilkent.edu.tr RETINA Pattern Recognition Tutorial, Summer 2005 Overview Statistical
More informationFully Automatic Methodology for Human Action Recognition Incorporating Dynamic Information
Fully Automatic Methodology for Human Action Recognition Incorporating Dynamic Information Ana González, Marcos Ortega Hortas, and Manuel G. Penedo University of A Coruña, VARPA group, A Coruña 15071,
More informationAnnouncements. Recognition. Recognition. Recognition. Recognition. Homework 3 is due May 18, 11:59 PM Reading: Computer Vision I CSE 152 Lecture 14
Announcements Computer Vision I CSE 152 Lecture 14 Homework 3 is due May 18, 11:59 PM Reading: Chapter 15: Learning to Classify Chapter 16: Classifying Images Chapter 17: Detecting Objects in Images Given
More informationMOTION CAPTURE DATA PROCESSING - MOTION EDITING / RETARGETING - MOTION CONTROL / GRAPH - INVERSE KINEMATIC. Alexandre Meyer Master Informatique
1 MOTION CAPTURE DATA PROCESSING - MOTION EDITING / RETARGETING - MOTION CONTROL / GRAPH - INVERSE KINEMATIC Alexandre Meyer Master Informatique Overview: Motion data processing In this course Motion editing
More informationBehaviour based particle filtering for human articulated motion tracking
Loughborough University Institutional Repository Behaviour based particle filtering for human articulated motion tracking This item was submitted to Loughborough University's Institutional Repository by
More informationTracking in Action Space
Tracking in Action Space Dennis L. Herzog, and Volker Krüger Copenhagen Institute of Technology, Aalborg University, Lautrupvang 2, 2750 Ballerup, Denmark {deh,vok}@cvmi.aau.dk Abstract. The recognition
More informationUnsupervised Human Members Tracking Based on an Silhouette Detection and Analysis Scheme
Unsupervised Human Members Tracking Based on an Silhouette Detection and Analysis Scheme Costas Panagiotakis and Anastasios Doulamis Abstract In this paper, an unsupervised, automatic video human members(human
More informationMachine Learning A W 1sst KU. b) [1 P] Give an example for a probability distributions P (A, B, C) that disproves
Machine Learning A 708.064 11W 1sst KU Exercises Problems marked with * are optional. 1 Conditional Independence I [2 P] a) [1 P] Give an example for a probability distribution P (A, B, C) that disproves
More informationTextural Features for Image Database Retrieval
Textural Features for Image Database Retrieval Selim Aksoy and Robert M. Haralick Intelligent Systems Laboratory Department of Electrical Engineering University of Washington Seattle, WA 98195-2500 {aksoy,haralick}@@isl.ee.washington.edu
More informationimages is then estimated using the function approximator, such as the neural networks, or by matching novel images to the examples in the database. Wi
From Gaze to Focus of Attention Rainer Stiefelhagen 1, Michael Finke 2, Jie Yang 2, and Alex Waibel 12 1 Universitat Karlsruhe, Computer Science, ILKD Am Fasanengarten 5, 76131 Karlsruhe, Germany stiefel@ira.uka.de
More informationEffect of Initial HMM Choices in Multiple Sequence Training for Gesture Recognition
Effect of Initial HMM Choices in Multiple Sequence Training for Gesture Recognition Nianjun Liu, Richard I.A. Davis, Brian C. Lovell and Peter J. Kootsookos Intelligent Real-Time Imaging and Sensing (IRIS)
More informationOverview. Animation is a big topic We will concentrate on character animation as is used in many games today. humans, animals, monsters, robots, etc.
ANIMATION Overview Animation is a big topic We will concentrate on character animation as is used in many games today humans, animals, monsters, robots, etc. Character Representation A character is represented
More informationMixture Models and EM
Mixture Models and EM Goal: Introduction to probabilistic mixture models and the expectationmaximization (EM) algorithm. Motivation: simultaneous fitting of multiple model instances unsupervised clustering
More informationReal-Time Human Pose Inference using Kernel Principal Component Pre-image Approximations
1 Real-Time Human Pose Inference using Kernel Principal Component Pre-image Approximations T. Tangkuampien and D. Suter Institute for Vision Systems Engineering Monash University, Australia {therdsak.tangkuampien,d.suter}@eng.monash.edu.au
More informationCLASSIFICATION WITH RADIAL BASIS AND PROBABILISTIC NEURAL NETWORKS
CLASSIFICATION WITH RADIAL BASIS AND PROBABILISTIC NEURAL NETWORKS CHAPTER 4 CLASSIFICATION WITH RADIAL BASIS AND PROBABILISTIC NEURAL NETWORKS 4.1 Introduction Optical character recognition is one of
More informationPattern Recognition. Kjell Elenius. Speech, Music and Hearing KTH. March 29, 2007 Speech recognition
Pattern Recognition Kjell Elenius Speech, Music and Hearing KTH March 29, 2007 Speech recognition 2007 1 Ch 4. Pattern Recognition 1(3) Bayes Decision Theory Minimum-Error-Rate Decision Rules Discriminant
More informationUnsupervised learning in Vision
Chapter 7 Unsupervised learning in Vision The fields of Computer Vision and Machine Learning complement each other in a very natural way: the aim of the former is to extract useful information from visual
More informationAssignment 2. Unsupervised & Probabilistic Learning. Maneesh Sahani Due: Monday Nov 5, 2018
Assignment 2 Unsupervised & Probabilistic Learning Maneesh Sahani Due: Monday Nov 5, 2018 Note: Assignments are due at 11:00 AM (the start of lecture) on the date above. he usual College late assignments
More informationClustering K-means. Machine Learning CSEP546 Carlos Guestrin University of Washington February 18, Carlos Guestrin
Clustering K-means Machine Learning CSEP546 Carlos Guestrin University of Washington February 18, 2014 Carlos Guestrin 2005-2014 1 Clustering images Set of Images [Goldberger et al.] Carlos Guestrin 2005-2014
More informationVideo Texture. A.A. Efros
Video Texture A.A. Efros 15-463: Computational Photography Alexei Efros, CMU, Fall 2005 Weather Forecasting for Dummies Let s predict weather: Given today s weather only, we want to know tomorrow s Suppose
More informationVerification: is that a lamp? What do we mean by recognition? Recognition. Recognition
Recognition Recognition The Margaret Thatcher Illusion, by Peter Thompson The Margaret Thatcher Illusion, by Peter Thompson Readings C. Bishop, Neural Networks for Pattern Recognition, Oxford University
More informationStatic Gesture Recognition with Restricted Boltzmann Machines
Static Gesture Recognition with Restricted Boltzmann Machines Peter O Donovan Department of Computer Science, University of Toronto 6 Kings College Rd, M5S 3G4, Canada odonovan@dgp.toronto.edu Abstract
More informationPerson identification from spatio-temporal 3D gait
200 International Conference on Emerging Security Technologies Person identification from spatio-temporal 3D gait Yumi Iwashita Ryosuke Baba Koichi Ogawara Ryo Kurazume Information Science and Electrical
More informationTowards the Automatic Analysis of Complex Human Body Motions. Abstract
Towards the Automatic Analysis of Complex Human Body Motions J. Rittscher A. Blake S. J. Roberts GE Global Research Microsoft Research University of Oxford 1 Research Circle 7 JJ Thompson Avenue Parks
More informationAn Object Detection System using Image Reconstruction with PCA
An Object Detection System using Image Reconstruction with PCA Luis Malagón-Borja and Olac Fuentes Instituto Nacional de Astrofísica Óptica y Electrónica, Puebla, 72840 Mexico jmb@ccc.inaoep.mx, fuentes@inaoep.mx
More informationInferring 3D Body Pose from Silhouettes using Activity Manifold Learning
CVPR 4 Inferring 3D Body Pose from Silhouettes using Activity Manifold Learning Ahmed Elgammal and Chan-Su Lee Department of Computer Science, Rutgers University, New Brunswick, NJ, USA {elgammal,chansu}@cs.rutgers.edu
More informationMobile Robotics. Mathematics, Models, and Methods. HI Cambridge. Alonzo Kelly. Carnegie Mellon University UNIVERSITY PRESS
Mobile Robotics Mathematics, Models, and Methods Alonzo Kelly Carnegie Mellon University HI Cambridge UNIVERSITY PRESS Contents Preface page xiii 1 Introduction 1 1.1 Applications of Mobile Robots 2 1.2
More informationMODELING HUMAN GAITS WITH SUBTLETIES. Alessandro Bissacco Payam Saisan Stefano Soatto
MODELING HUMAN GAITS WITH SUBTLETIES Alessandro Bissacco Payam Saisan Stefano Soatto Department of Computer Science, University of California, Los Angeles, CA 90095 {bissacco,soatto}@cs.ucla.edu Department
More informationClustering Documents in Large Text Corpora
Clustering Documents in Large Text Corpora Bin He Faculty of Computer Science Dalhousie University Halifax, Canada B3H 1W5 bhe@cs.dal.ca http://www.cs.dal.ca/ bhe Yongzheng Zhang Faculty of Computer Science
More informationAnnouncements. Recognition I. Gradient Space (p,q) What is the reflectance map?
Announcements I HW 3 due 12 noon, tomorrow. HW 4 to be posted soon recognition Lecture plan recognition for next two lectures, then video and motion. Introduction to Computer Vision CSE 152 Lecture 17
More informationBoosting Sex Identification Performance
Boosting Sex Identification Performance Shumeet Baluja, 2 Henry Rowley shumeet@google.com har@google.com Google, Inc. 2 Carnegie Mellon University, Computer Science Department Abstract This paper presents
More information